Classification of two species of Bidens based on discrete stationary wavelet transform extraction of FTIR spectra combined with probability neural network

Horizontal attenuated total reflectance Fourier transform infrared spectroscopy (HATR-FTIR) technique is used to measure the FTIR spectra of Bidens frondosa L. seed and Bidens bipinnata L. seed. Because they belong to the same genus plants, their chemical components are very similar. In order to extract the difference between B. frondosa L. seed and B. bipinnata L. seed, discrete stationary wavelet transform (DSWT) is used to decompose the FTIR spectra of B. frondosa L. seed and B. bipinnata L. seed. Three main scales are selected as the feature extracting space in the DSWT domain. According to the distribution of FTIR spectra of B. frondosa L. seed and B. bipinnata L. seed, three feature regions are determined by the information content of substance construction in their FTIR spectra in the DSWT domain. Thus nine feature parameters form the feature vector. The feature vector is input to the probability neural network (PNN) to train so as to accurately classify the B. frondosa L. seed and B. bipinnata L. seed. Experimental results show that the accurate recognition rate between B. frondosa L. seed and B. bipinnata L. seed is in the range of 98–100% by using the proposed method.